rtnetmon/main.go

// netmon – dual-interface network dashboard (curses)
// WTFPL – 2025-05-16  sneak@sneak.berlin
package main

import (
	"context"
	"encoding/json"
	"flag"
	"fmt"
	"math"
	"net"
	"os"
	"os/exec"
	"os/signal"
	"strconv"
	"strings"
	"sync"
	"syscall"
	"time"

	"github.com/gdamore/tcell/v2"
)

/*────────────────────────  CLI  ────────────────────────*/

var (
	ifaceA = flag.String("ifaceA", "gu0", "primary network interface")
	labelA = flag.String("labelA", "gu LAN - VPN outbound", "label for ifaceA")
	ifaceB = flag.String("ifaceB", "backhaul0", "secondary network interface")
	labelB = flag.String("labelB", "Cox cable direct", "label for ifaceB")

	hostCSV = flag.String("hosts",
		"1.1.1.1,8.8.8.8,8.8.4.4,google.com,github.com,"+
			"console.aws.amazon.com,console.cloud.google.com,"+
			"fast.com,cloudflare.com,datavi.be",
		"hosts for reachability checks")
)

/*──────────────────────── constants ─────────────────────*/

const (
	icmpTimeout      = 500 * time.Millisecond
	tcpTimeout       = 500 * time.Millisecond
	packetLossPings  = 20
	packetLossPeriod = 5 * time.Second
	statsHistory     = 300               // 5 min @1 Hz
	screenRefresh    = 500 * time.Millisecond
)

/*──────────────────────── runtime data ──────────────────*/

type InterfaceStatus struct {
	Name, Label, IPInfo       string
	Reachable                 map[string]bool
	Loss                      map[string]float64
	TCP                       map[string][]float64
	TotalICMPReq, TotalICMPRep int
	DroppedCount              int
	LastDrop, LastPing        time.Time
	SpinFrame                 int
	mu                        sync.RWMutex
}

/*──────────────────────── colours ───────────────────────*/

var (
	cBrightGreen = tcell.StyleDefault.Foreground(tcell.ColorGreen).Bold(true)
	cGreen       = tcell.StyleDefault.Foreground(tcell.ColorGreen)
	cYellow      = tcell.StyleDefault.Foreground(tcell.ColorYellow)
	cRed         = tcell.StyleDefault.Foreground(tcell.ColorRed)
	cBrightRed   = tcell.StyleDefault.Foreground(tcell.ColorRed).Bold(true)
	cDefault     = tcell.StyleDefault
)

func styleLatency(ms float64) tcell.Style {
	switch {
	case ms < 50:
		return cBrightGreen
	case ms < 100:
		return cGreen
	case ms < 200: // new threshold
		return cYellow
	default:
		return cRed
	}
}

func styleLoss(pct float64) tcell.Style {
	switch {
	case pct == 0:
		return cBrightGreen
	case pct < 5:
		return cYellow
	default:
		return cBrightRed
	}
}

/*──────────────────────── maths ─────────────────────────*/

func minMaxAvgStd(xs []float64) (min, max, avg, std float64) {
	if len(xs) == 0 {
		return
	}
	min, max = xs[0], xs[0]
	var sum float64
	for _, v := range xs {
		if v < min {
			min = v
		}
		if v > max {
			max = v
		}
		sum += v
	}
	avg = sum / float64(len(xs))
	var vs float64
	for _, v := range xs {
		d := v - avg
		vs += d * d
	}
	std = math.Sqrt(vs / float64(len(xs)))
	return
}

/*──────────────────────── external lookup ───────────────*/

type ipInfoResp struct{ IP, Hostname, Org string }

func fetchIPInfo(iface string) string {
	ctx, cancel := context.WithTimeout(context.Background(), 2*time.Second)
	defer cancel()
	out, err := exec.CommandContext(ctx, "curl", "-s", "--interface", iface, "--max-time", "2", "ipinfo.io").Output()
	if err != nil {
		return "(ipinfo error)"
	}
	var r ipInfoResp
	_ = json.Unmarshal(out, &r)
	if r.IP == "" {
		return "(ipinfo parse error)"
	}
	return fmt.Sprintf("%s [%s] %s", r.IP, r.Hostname, r.Org)
}

/*──────────────────────── ICMP helpers ──────────────────*/

func pingOnce(iface, host string) bool {
	ctx, cancel := context.WithTimeout(context.Background(), icmpTimeout)
	defer cancel()
	return exec.CommandContext(ctx, "ping", "-I", iface, "-c1", "-W1", host).Run() == nil
}

func lossPercent(iface, host string) float64 {
	ctx, cancel := context.WithTimeout(context.Background(), 3*time.Second)
	defer cancel()
	out, err := exec.CommandContext(ctx, "ping", "-q", "-i", "0.05",
		"-c", fmt.Sprint(packetLossPings), "-W1", "-I", iface, host).CombinedOutput()
	if err != nil {
		return 1.0
	}
	for _, ln := range strings.Split(string(out), "\n") {
		if strings.Contains(ln, "packet loss") {
			for _, f := range strings.Fields(ln) {
				if strings.HasSuffix(f, "%") {
					p, _ := strconv.ParseFloat(strings.TrimSuffix(f, "%"), 64)
					return p / 100.0
				}
			}
		}
	}
	return 1.0
}

/*──────────────────────── TCP helpers ───────────────────*/

func localAddr(iface string) (net.Addr, error) {
	ni, err := net.InterfaceByName(iface)
	if err != nil {
		return nil, err
	}
	for _, a := range func() []net.Addr { a, _ := ni.Addrs(); return a }() {
		if ipnet, ok := a.(*net.IPNet); ok && ipnet.IP.To4() != nil {
			return &net.TCPAddr{IP: ipnet.IP}, nil
		}
	}
	return nil, fmt.Errorf("no IPv4 on %s", iface)
}

func tcpDuration(iface, hp string) time.Duration {
	la, err := localAddr(iface)
	if err != nil {
		return tcpTimeout
	}
	d := net.Dialer{Timeout: tcpTimeout, LocalAddr: la}
	start := time.Now()
	c, err := d.Dial("tcp", hp)
	if err != nil {
		return tcpTimeout
	}
	c.Close()
	return time.Since(start)
}

/*──────────────────────── spinner ───────────────────────*/

var spins = []rune{'|', '/', '-', '\\'}

func (st *InterfaceStatus) spin() { st.SpinFrame = (st.SpinFrame + 1) % len(spins) }

/*──────────────────────── screen helpers ────────────────*/

func put(scr tcell.Screen, x, y int, txt string, st tcell.Style) {
	for i, r := range txt {
		scr.SetContent(x+i, y, r, nil, st)
	}
}

func hline(w int) string { return strings.Repeat("=", w) }

/*──────────────────────── UI draw helpers ───────────────*/

func ifaceHeaderStyle(sta *InterfaceStatus) tcell.Style {
	sta.mu.RLock()
	defer sta.mu.RUnlock()
	if sta.DroppedCount > 0 {
		return cBrightRed
	}
	for _, ok := range sta.Reachable {
		if !ok {
			return cBrightRed
		}
	}
	for _, lp := range sta.Loss {
		if lp > 0 {
			return cBrightRed
		}
	}
	return cBrightGreen
}

const (
	hostCol = 0
	minCol  = 18
	avgCol  = 27
	maxCol  = 36
	stdCol  = 45
	nCol    = 56
)

func drawLatencyRow(scr tcell.Screen, y int, host string, mi, av, ma, sd float64, n int) {
	put(scr, hostCol, y, fmt.Sprintf("%-17s", host), cDefault)
	put(scr, minCol, y, fmt.Sprintf("%6.0fms", mi), styleLatency(mi))
	put(scr, avgCol, y, fmt.Sprintf("%6.0fms", av), styleLatency(av))
	put(scr, maxCol, y, fmt.Sprintf("%6.0fms", ma), styleLatency(ma))
	put(scr, stdCol, y, fmt.Sprintf("%7.0fms", sd), cDefault)
	put(scr, nCol, y, fmt.Sprintf("%5d", n), cDefault)
}

func drawIface(scr tcell.Screen, y, w int, sta *InterfaceStatus) int {
	/* header */
	put(scr, 0, y, hline(w), cDefault)
	hs := ifaceHeaderStyle(sta)
	sta.mu.RLock()
	head := fmt.Sprintf("== %c %s —     %s", spins[sta.SpinFrame], sta.Label, sta.IPInfo)
	sta.mu.RUnlock()
	if len(head) > w {
		head = head[:w]
	}
	put(scr, 0, y+1, head, hs)
	put(scr, 0, y+2, hline(w), cDefault)
	y += 4

	/* reachability */
	sta.mu.RLock()
	total := len(sta.Reachable)
	good := 0
	for _, ok := range sta.Reachable {
		if ok {
			good++
		}
	}
	rStyle := cBrightGreen
	if good != total {
		rStyle = cBrightRed
	}
	age := time.Since(sta.LastPing).Round(time.Second)
	put(scr, 0, y, fmt.Sprintf("Reachable: %d/%d (at %s, age %s)", good, total, sta.LastPing.Format("15:04:05"), age), rStyle)
	y++
	if good == total {
		put(scr, 0, y, "Unreachable: none", cDefault)
	} else {
		var down []string
		for h, ok := range sta.Reachable {
			if !ok {
				down = append(down, h)
			}
		}
		put(scr, 0, y, "Unreachable: "+strings.Join(down, ", "), cBrightRed)
	}
	y += 2

	/* packet loss */
	put(scr, 0, y, "Packet Loss:", cDefault)
	y++
	for _, h := range packetLossHosts {
		pct := sta.Loss[h] * 100
		put(scr, 0, y, fmt.Sprintf("%-16s %5.0f%%", h+":", pct), styleLoss(pct))
		y++
	}
	dAge := "N/A"
	if !sta.LastDrop.IsZero() {
		dAge = time.Since(sta.LastDrop).Round(time.Second).String()
	}
	put(scr, 0, y, fmt.Sprintf("Dropped: %d (last at %s, age %s)", sta.DroppedCount, sta.LastDrop.Format("15:04:05"), dAge), cDefault)
	y += 2

	/* TCP table */
	put(scr, 0, y, "TCP Connect Stats:", cDefault)
	y++
	put(scr, hostCol, y, "Host", cDefault)
	put(scr, minCol, y, "  min", cDefault)
	put(scr, avgCol, y, "  avg", cDefault)
	put(scr, maxCol, y, "  max", cDefault)
	put(scr, stdCol, y, " stddev", cDefault)
	put(scr, nCol, y, "   n", cDefault)
	y++

	for _, hp := range tcpTestHosts {
		hist := sta.TCP[hp]
		if len(hist) == 0 {
			continue
		}
		mi, ma, av, sd := minMaxAvgStd(hist)
		drawLatencyRow(scr, y, hp, mi, av, ma, sd, len(hist))
		y++
	}
	y++

	/* totals */
	put(scr, 0, y, fmt.Sprintf("Total ICMP Requests: %d", sta.TotalICMPReq), cDefault)
	y++
	put(scr, 0, y, fmt.Sprintf("Total ICMP Replies:  %d", sta.TotalICMPRep), cDefault)
	y += 2
	sta.mu.RUnlock()
	return y
}

/*──────────────────────── loops ────────────────────────*/

func reachLoop(ctx context.Context, st *InterfaceStatus, hosts []string) {
	tk := time.NewTicker(time.Second)
	defer tk.Stop()
	for {
		select {
		case <-ctx.Done():
			return
		case <-tk.C:
			var wg sync.WaitGroup
			res := make(map[string]bool, len(hosts))
			mu := sync.Mutex{}
			for _, h := range hosts {
				wg.Add(1)
				go func(host string) {
					defer wg.Done()
					st.mu.Lock()
					st.TotalICMPReq++
					st.mu.Unlock()

					ok := pingOnce(st.Name, host)
					mu.Lock()
					res[host] = ok
					mu.Unlock()

					st.mu.Lock()
					if ok {
						st.TotalICMPRep++
						st.spin()
					} else {
						st.DroppedCount++
						st.LastDrop = time.Now()
					}
					st.mu.Unlock()
				}(h)
			}
			wg.Wait()
			st.mu.Lock()
			st.Reachable = res
			st.LastPing = time.Now()
			st.mu.Unlock()
		}
	}
}

func lossLoop(ctx context.Context, st *InterfaceStatus, hosts []string) {
	tk := time.NewTicker(packetLossPeriod)
	defer tk.Stop()
	for {
		select {
		case <-ctx.Done():
			return
		case <-tk.C:
			var wg sync.WaitGroup
			res := make(map[string]float64, len(hosts))
			mu := sync.Mutex{}
			for _, h := range hosts {
				wg.Add(1)
				go func(host string) {
					defer wg.Done()
					lp := lossPercent(st.Name, host)
					mu.Lock()
					res[host] = lp
					if lp == 0 {
						st.spin()
					}
					mu.Unlock()
				}(h)
			}
			wg.Wait()
			st.mu.Lock()
			for k, v := range res {
				st.Loss[k] = v
			}
			st.mu.Unlock()
		}
	}
}

func tcpLoop(ctx context.Context, st *InterfaceStatus, hosts []string) {
	tk := time.NewTicker(time.Second)
	defer tk.Stop()
	for {
		select {
		case <-ctx.Done():
			return
		case <-tk.C:
			for _, hp := range hosts {
				ms := float64(tcpDuration(st.Name, hp).Milliseconds())
				st.mu.Lock()
				if ms < float64(tcpTimeout.Milliseconds()) {
					st.spin()
				}
				hist := st.TCP[hp]
				if len(hist) >= statsHistory {
					hist = hist[1:]
				}
				st.TCP[hp] = append(hist, ms)
				st.mu.Unlock()
			}
		}
	}
}

/*──────────────────────── UI ───────────────────────────*/

func uiLoop(ctx context.Context, scr tcell.Screen, a, b *InterfaceStatus, start time.Time) {
	defer scr.Fini()
	tk := time.NewTicker(screenRefresh)
	defer tk.Stop()

	spin := 0
	for {
		select {
		case <-ctx.Done():
			return
		case <-tk.C:
			w, _ := scr.Size()
			scr.Clear()
			put(scr, 0, 0, hline(w), cDefault)
			put(scr, 0, 1, fmt.Sprintf("== %c %s", spins[spin%len(spins)], time.Now().Format(time.RFC1123Z)), cDefault)
			put(scr, 0, 2, hline(w), cDefault)
			spin++
			put(scr, 0, 3, "Runtime: "+time.Since(start).Round(time.Second).String(), cDefault)
			y := 5
			y = drawIface(scr, y, w, a)
			_ = drawIface(scr, y, w, b)
			scr.Show()
		}
	}
}

/*──────────────────────── main ─────────────────────────*/

var (
	reachHosts      []string
	packetLossHosts = []string{"github.com", "google.com", "1.1.1.1", "8.8.8.8"}
	tcpTestHosts    = []string{
		"datavi.be:443", "fast.com:443", "cloudflare.com:443",
		"console.aws.amazon.com:443", "console.cloud.google.com:443",
	}
)

func main() {
	flag.Parse()
	reachHosts = strings.Split(*hostCSV, ",")

	newStatus := func(name, label string) *InterfaceStatus {
		return &InterfaceStatus{
			Name:      name,
			Label:     label,
			IPInfo:    fetchIPInfo(name),
			Reachable: map[string]bool{},
			Loss:      map[string]float64{},
			TCP:       map[string][]float64{},
		}
	}
	a, b := newStatus(*ifaceA, *labelA), newStatus(*ifaceB, *labelB)

	screen, err := tcell.NewScreen()
	if err != nil {
		panic(err)
	}
	if err = screen.Init(); err != nil {
		panic(err)
	}

	ctx, cancel := context.WithCancel(context.Background())
	defer cancel()

	// terminal Ctrl-C / SIGTERM
	sig := make(chan os.Signal, 1)
	signal.Notify(sig, os.Interrupt, syscall.SIGTERM)
	go func() { <-sig; cancel() }()

	// in-app Ctrl-C / q
	go func() {
		for {
			ev := screen.PollEvent()
			if ev == nil {
				return
			}
			switch e := ev.(type) {
			case *tcell.EventKey:
				if e.Key() == tcell.KeyCtrlC || e.Rune() == 'q' {
					cancel()
					return
				}
			}
		}
	}()

	go reachLoop(ctx, a, reachHosts)
	go reachLoop(ctx, b, reachHosts)
	go lossLoop(ctx, a, packetLossHosts)
	go lossLoop(ctx, b, packetLossHosts)
	go tcpLoop(ctx, a, tcpTestHosts)
	go tcpLoop(ctx, b, tcpTestHosts)

	uiLoop(ctx, screen, a, b, time.Now())
}